🧠 What Claude Remembers About You (And How To Edit It)

Your supermarket loyalty card knows what time you shop, what you buy when you’re stressed, and roughly when your washing machine broke. It has never once shown you the file.

Claude’s memory - the feature people find creepiest on paper - is the opposite arrangement. It’s a notebook, you’re allowed to read it, and you hold the pen. You can see every note, correct any note, and tear out pages whenever you like. Of all the things software quietly knows about you, this is the rare one with a “show me” button.

This guide demystifies the whole thing: what actually gets stored, how to look at it, how to change it, how to switch it off, and the one small habit that turns memory from a nice-to-have into the reason Claude feels like it gets you. I run everything through four verbs: Read, Edit, Bin, Feed.

First, thirty seconds of what memory actually is. When memory’s on (it’s free for everyone - Settings → Capabilities if you want to check), Claude keeps short notes on the useful stuff you tell it: what you do, ongoing projects, preferences, the fact you hate bullet points. Notes, not recordings. It’s the difference between a colleague who remembers “Sam’s daughter starts secondary school in September” and one who has transcripts of your every conversation. Claude’s the first colleague.

Verb 1: Read it

What it is: looking at the actual notebook, so memory stops being an imagined dossier and becomes a slightly mundane list.

How to do it: open a chat and type: “What do you remember about me? List everything.” That’s genuinely the mechanism - no hidden database to hunt down. There’s also a memory view in the settings if you prefer poking around menus (they move about occasionally, so if it’s not where you expect, it’ll be nearby), but the chat question always works and gives you the friendlier version.

The bit nobody tells you: do this before deciding how you feel about memory. Almost everyone I teach imagines something far more sinister than what appears, which is usually fifteen lines of things like “runs a dog-grooming business, prefers short answers, planning a trip to Portugal”. Reading it is the fastest cure for the creeps. It’s not a dossier… it’s a Post-it note.

Verb 2: Edit it

What it is: correcting the notebook when it’s wrong or stale - because a memory that’s out of date is worse than no memory at all.

How to do it: in plain sentences, in any chat. “Update your memory: I’ve moved from Leeds to Sheffield.” “You’ve got my job title wrong, it’s office manager now.” “Stop noting that I’m new to AI, I’d like normal-speed answers these days.” Claude confirms the change and the notebook moves on.

Say you told Claude all about your wedding-cake business back in March, then pivoted to bookkeeping over the summer. Without an edit, October’s chats keep gently steering everything back to buttercream. One sentence - “update your memory: I closed the cake business, I do bookkeeping for tradespeople now” - and every future conversation starts in the right place.

The bit nobody tells you: stale memories are behind most “Claude’s gone a bit weird lately” complaints. It isn’t broken; it’s working faithfully from old notes. If answers keep leaning on a life you no longer live, edit the notebook before you blame the assistant.

Verb 3: Bin it

What it is: the delete end of the spectrum - from removing one note to switching the whole thing off.

How to do it: three levels, smallest first. For one memory: “Forget what I told you about my salary negotiation.” Done, gone. For a proper clear-out: use the spring-clean prompt below. For the lot: Settings → Capabilities, switch memory off, and Claude goes back to treating every conversation as a fresh introduction. No penalty, no guilt trip, reversible whenever.

The bit nobody tells you: memory and training data are two entirely separate dials, and people constantly blur them. Memory (Capabilities) is Claude’s notebook for your benefit, in your conversations. Whether chats help improve future models is a different toggle in Settings → Privacy (sometimes called Data Controls), and you can set each independently. Wanting Claude to remember you does not oblige you to share anything, and vice versa. Knowing they’re separate is half the reassurance.

Verb 4: Feed it

What it is: the habit that makes memory brilliant instead of merely fine - deliberately telling Claude to remember decisions, not just letting it passively collect facts.

How to do it: when a conversation lands somewhere useful, end it with one line: “Remember this decision: my day rate is £180 and I’m done discounting for friends-of-friends.” “Remember: emails always get signed off ‘Cheers, Priya’, never ‘Kind regards’.” “Remember we chose the blue branding and stop suggesting alternatives.”

The bit nobody tells you: left to its own devices, memory hoovers up biography - who you are, what you do. Useful, but the gold is rulings: prices you’ve settled, tones you’ve chosen, boundaries you’ve drawn, options you’ve closed. Feed it decisions and Claude stops relitigating things you settled weeks ago - it just quietly applies them, which is when people start saying it reads their mind. It doesn’t. You fed it the answers.

The spring-clean

Do this every couple of months, or whenever life shifts - new job, house move, business pivot. Ten minutes, oddly satisfying, like clearing out a drawer.

Nick this one:

Time for a memory spring-clean.

1. List everything you currently remember about me, grouped under: who I am, work, preferences, ongoing projects and plans, and anything else.
2. Mark anything that looks out of date, too vague to be useful, or that you're not confident about.
3. Take me through the marked items one at a time and ask: keep, update, or forget? Wait for my answer each time.
4. When we've finished, show me the tidied final version, and tell me if there's anything worth remembering about me that ISN'T in there yet.

That last line earns its keep - it regularly surfaces a decision you made mid-chat weeks ago that never got properly filed.

Where this stops working

• Memory is notes, not transcripts. Claude won’t recall the exact wording of something from six weeks ago out of its notebook. That’s what “search and reference chats” (same Capabilities page) is for - it lets Claude look back through old conversations when it’s relevant. The two work as a pair: notebook for the gist, search for the receipts.
• It can’t remember what you never said. No note-swapping with your email, your calendar, or your browser history. Sparse chats mean a sparse notebook, and that’s the system working as designed.
• It won’t judge what deserves remembering as well as you can. It makes decent guesses, but Verb 4 exists because you know which decisions matter.

One thing, then close the tab

Ask the question. Today, one line: “What do you remember about me?” Whatever your current feeling about AI memory, five minutes from now it’ll be based on the actual contents of the actual notebook, and you’ll know exactly how to change anything you don’t like.

One thing memory should never have to hold: passwords, bank details, or other people’s private business - Never Paste These Into AI covers exactly where that line sits. And if you want your whole setup this well-tuned, session by session with a playbook for your exact job, that’s what The Claude Kickstart is for (coming soon).